\nThe model used wasn't made by me.\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F10bda72d118451c12f01b66c34b77424%2Fd005a3526211fa9e6a68082a669a51e1.png?generation=1574531154433068&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F8f9b62bb98b1bee4914fc681bbe427a4%2F97ccbfac05234cd529227cbcf7f247be.png?generation=1574531171345504&alt=media\")
An interesting way to estimate depth, isn't very accurate though. Green dots are from train.csv and predictions are made by converting x,y,z from train.csv into image pixel coordinates and indexing the model output.
\nThe model used wasn't made by me.\n\n
An interesting way to estimate depth, isn't very accurate though. Green dots are from train.csv and predictions are made by converting x,y,z from train.csv into image pixel coordinates and indexing the model output.
\nThe model used wasn't made by me.\n\n
Awesome !
", "rawMarkdown": "Awesome !", "votes": null }, { "id": "679986", "postDate": "11/23/2019 19:19:34", "content": "looks nice. Which model was used for the depth prediction?
", "rawMarkdown": "looks nice. Which model was used for the depth prediction?", "votes": null }, { "id": "679996", "postDate": "11/23/2019 19:41:34", "content": "https://github.com/nianticlabs/monodepth2
", "rawMarkdown": "https://github.com/nianticlabs/monodepth2", "votes": null }, { "id": "709458", "postDate": "01/03/2020 14:38:36", "content": "Hi, GreatGameDota:\nwould you mind share the code you use to get depth map? thanks๐ @greatgamedota
", "rawMarkdown": "Hi, GreatGameDota:\nwould you mind share the code you use to get depth map? thanks๐ @greatgamedota", "votes": null }, { "id": "709591", "postDate": "01/03/2020 17:40:05", "content": "Here's the colab notebook, tell me if the link works: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7
", "rawMarkdown": "Here's the colab notebook, tell me if the link works: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7", "votes": null }, { "id": "709878", "postDate": "01/04/2020 03:20:52", "content": "Thanks, and it needs your permission, or would you mind sharing .ipynb document? I'm using colab too.๐ \nHere's a algorithm that uses depth map in a better way. https://arxiv.org/abs/1812.07179 @greatgamedota \n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3876174%2F62aa13a4ad93f0ddcd8aaf6d96796f8b%2F2020-01-04_111625.png?generation=1578107849054627&alt=media\")
Whoops, here's the sharable link: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7
\nThanks for the link, I'll check it out!
Hi, I see the outputs have 4 disp, what is it used for? I plot it out, they are same. @greatgamedota
\n\ndisp = outputs[(\"disp\", 0)]
", "rawMarkdown": "Hi, I see the outputs have 4 disp, what is it used for? I plot it out, they are same. @greatgamedota \n\ndisp = outputs[(\"disp\", 0)]", "votes": null }, { "id": "710279", "postDate": "01/04/2020 14:33:57", "content": "It's just the same output but at different resolutions, you can control it when initializing the decoder: depth_decoder = networks.DepthDecoder(num_ch_enc=encoder.num_ch_enc, scales=range(4))
If your still working with this I found a much better depth predictor.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F9aed3d8ff36f8c054dd8afca31f25e6e%2Fvnl.png?generation=1578244717716386&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3543139%2F1466629fc162dc5eb910a031d3c01883%2Fpred_vnl.png?generation=1578248372851148&alt=media\")
Thanks, I havn't try it, the timeline is coming, no time to try new model.๐
", "rawMarkdown": "Thanks, I havn't try it, the timeline is coming, no time to try new model.๐", "votes": null } ], "comments": [ { "id": 679944, "author_name": "phoenix9032", "author_url": "", "post_date": "11/23/2019 17:49:55", "content": "Awesome !
", "votes": null, "replies": [] }, { "id": 679986, "author_name": "tugstugi", "author_url": "", "post_date": "11/23/2019 19:19:34", "content": "looks nice. Which model was used for the depth prediction?
", "votes": null, "replies": [ { "id": 679996, "author_name": "greatgamedota", "author_url": "", "post_date": "11/23/2019 19:41:34", "content": "https://github.com/nianticlabs/monodepth2
", "votes": null, "replies": [] } ] }, { "id": 709458, "author_name": "diegojohnson", "author_url": "", "post_date": "01/03/2020 14:38:36", "content": "Hi, GreatGameDota:\nwould you mind share the code you use to get depth map? thanks๐ @greatgamedota
", "votes": null, "replies": [ { "id": 709591, "author_name": "greatgamedota", "author_url": "", "post_date": "01/03/2020 17:40:05", "content": "Here's the colab notebook, tell me if the link works: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7
", "votes": null, "replies": [] }, { "id": 709878, "author_name": "diegojohnson", "author_url": "", "post_date": "01/04/2020 03:20:52", "content": "Thanks, and it needs your permission, or would you mind sharing .ipynb document? I'm using colab too.๐ \nHere's a algorithm that uses depth map in a better way. https://arxiv.org/abs/1812.07179 @greatgamedota \n
Whoops, here's the sharable link: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7
\nThanks for the link, I'll check it out!
Hi, I see the outputs have 4 disp, what is it used for? I plot it out, they are same. @greatgamedota
\n\ndisp = outputs[(\"disp\", 0)]
", "votes": null, "replies": [] }, { "id": 710279, "author_name": "greatgamedota", "author_url": "", "post_date": "01/04/2020 14:33:57", "content": "It's just the same output but at different resolutions, you can control it when initializing the decoder: depth_decoder = networks.DepthDecoder(num_ch_enc=encoder.num_ch_enc, scales=range(4))
If your still working with this I found a much better depth predictor.
\n\n
Thanks, I havn't try it, the timeline is coming, no time to try new model.๐
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "679942": "An interesting way to estimate depth, isn't very accurate though. Green dots are from train.csv and predictions are made by converting x,y,z from train.csv into image pixel coordinates and indexing the model output. \nThe model used wasn't made by me.\n\n", "679944": "Awesome !", "679986": "looks nice. Which model was used for the depth prediction?", "679996": "https://github.com/nianticlabs/monodepth2", "709458": "Hi, GreatGameDota:\nwould you mind share the code you use to get depth map? thanks๐ @greatgamedota", "709591": "Here's the colab notebook, tell me if the link works: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7", "709878": "Thanks, and it needs your permission, or would you mind sharing .ipynb document? I'm using colab too.๐ \nHere's a algorithm that uses depth map in a better way. https://arxiv.org/abs/1812.07179 @greatgamedota \n", "709880": "Whoops, here's the sharable link: https://colab.research.google.com/drive/1u5CynV40wqJRuf0kIWuK-aLSVzjjjMr7 \nThanks for the link, I'll check it out!", "710245": "Hi, I see the outputs have 4 disp, what is it used for? I plot it out, they are same. @greatgamedota \n\ndisp = outputs[(\"disp\", 0)]", "710279": "It's just the same output but at different resolutions, you can control it when initializing the decoder: ```depth_decoder = networks.DepthDecoder(num_ch_enc=encoder.num_ch_enc, scales=range(4))```", "711094": "If your still working with this I found a much better depth predictor. \n\n", "711398": "Thanks, I havn't try it, the timeline is coming, no time to try new model.๐" }, "source": "meta" }